Image segregation system with method for handling textures

Abstract

In an exemplary embodiment of the present invention, an automated, computerized method is provided for processing an image. According to a feature of the present invention, the method comprises the steps of converting a color band representation of the image to a homogeneous representation of spectral and spatial characteristics of a texture region in the image and utilizing the homogeneous representation of spectral and spatial characteristics of a texture region in the image to identify homogeneous tokens in the image.

Description

This application is a continuation-in-part of application Ser. No. 11/888,453, filed Aug. 1, 2007, entitled System And Method For Identifying Complex Tokens In An Image.

BACKGROUND OF THE INVENTION

A challenge in the utilization of computers to accurately and correctly perform operations relating to images is the development of algorithms that truly reflect and represent physical phenomena occurring in the visual world. For example, the ability of a computer to correctly and accurately distinguish between a shadow and a material object edge within an image has been a persistent challenge to scientists. An early and conventional approach to object edge detection involves an analysis of brightness boundaries in an image. In the analysis it is assumed that a boundary caused by a material object will be sharp, while a boundary caused by a shadow will be soft or gradual due to the penumbra effect of shadows. While this approach can be implemented by algorithms that can be accurately executed by a computer, the results will often be incorrect. In the real world there are many instances wherein shadows form sharp boundaries, and conversely, material object edges form soft boundaries. Thus, when utilizing conventional techniques for shadow and object edge recognition, there are significant possibilities for false positives and false negatives for shadow recognition. That is, for example, a material edge that imitates a shadow and is thus identified incorrectly by a computer as a shadow or a sharp shadow boundary that is incorrectly interpreted as an object boundary. Accordingly, there is a persistent need for the development of accurate and correct techniques that can be utilized in the operation of computers relating to images, to, for example, identify material and illumination characteristics of the image.

SUMMARY OF THE INVENTION

The present invention provides a method and system comprising image processing techniques that utilize spatio-spectral information relevant to an image to accurately and correctly identify illumination and material aspects of the image.

In a first exemplary embodiment of the present invention, an automated, computerized method is provided for processing an image. According to a feature of the present invention, the method comprises the steps of converting a color band representation of the image to an intensity histogram representation and utilizing the intensity histogram representation to identify homogeneous tokens in the image.

In a second exemplary embodiment of the present invention, an automated, computerized method is provided for processing an image. According to a feature of the present invention, the method comprises the steps of converting a color band representation of the image to a texton representation and utilizing the texton representation to identify homogeneous tokens in the image.

In a third exemplary embodiment of the present invention, an automated, computerized method is provided for processing an image. According to a feature of the present invention, the method comprises the steps of converting a color band representation of the image to a homogeneous representation of spectral and spatial characteristics of a texture region in the image and utilizing the homogeneous representation of spectral and spatial characteristics of a texture region in the image to identify homogeneous tokens in the image.

In accordance with yet further embodiments of the present invention, computer systems are provided, which include one or more computers configured (e.g., programmed) to perform the methods described above. In accordance with other embodiments of the present invention, computer readable media are provided which have stored thereon computer executable process steps operable to control a computer(s) to implement the embodiments described above. The automated, computerized methods can be performed by a digital computer, analog computer, optical sensor, state machine, sequencer or any device or apparatus that can be designed or programmed to carry out the steps of the methods of the present invention.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is a block diagram of a computer system arranged and configured to perform operations related to images.

FIG. 2a shows an n×m pixel array image file for an image stored in the computer system of FIG. 1.

FIG. 2b is an 11×11 pixel array formed from the upper left hand corner of the image file of FIG. 2a, for use in the generation of an histogram, according to a feature of the present invention.

FIG. 2c shows a diagram of the histogram generated from the 11×11 pixel array of FIG. 2b.

FIG. 3a is a flow chart for identifying Type C token regions in the image file of FIG. 2a, according to a feature of the present invention.

FIG. 3b is an original image used as an example in the identification of Type C tokens.

FIG. 3c shows Type C token regions in the image of FIG. 3b.

FIG. 3d shows Type B tokens, generated from the Type C tokens of FIG. 3c, according to a feature of the present invention.

FIG. 4 is a flow chart for a routine to test Type C tokens identified by the routine of the flow chart of FIG. 3a, according to a feature of the present invention.

FIG. 5 is a flow chart for constructing Type B tokens via an arbitrary boundary removal technique, according to a feature of the present invention.

FIG. 6 is a flow chart for creating a token graph, containing token map information, according to a feature of the present invention.

FIG. 7 is a flow chart for constructing Type B tokens via an adjacent planar token merging technique, according to a feature of the present invention.

FIG. 8 is a flow chart for generating Type C tokens via a local token analysis technique, according to a feature of the present invention.

FIG. 9 is a flow chart for constructing Type B tokens from Type C tokens generated via the local token analysis technique of FIG. 8, according to a feature of the present invention.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

Referring now to the drawings, and initially to FIG. 1, there is shown a block diagram of a computer system 10 arranged and configured to perform operations related to images. A CPU 12 is coupled to a device such as, for example, a digital camera 14 via, for example, a USB port. The digital camera 14 operates to download images stored locally on the camera 14, to the CPU 12. The CPU 12 stores the downloaded images in a memory 16 as image files 18. The image files 18 can be accessed by the CPU 12 for display on a monitor 20, or for print out on a printer 22.

Alternatively, the CPU 12 can be implemented as a microprocessor embedded in a device such as, for example, the digital camera 14 or a robot. The CPU 12 can also be equipped with a real time operating system for real time operations related to images, in connection with, for example, a robotic operation or an interactive operation with a user.

As shown in FIG. 2a, each image file 18 comprises an n×m pixel array. Each pixel, p, is a picture element corresponding to a discrete portion of the overall image. All of the pixels together define the image represented by the image file 18. Each pixel comprises a digital value corresponding to a set of color bands, for example, red, green and blue color components (RGB) of the picture element. The present invention is applicable to any multi-band image, where each band corresponds to a piece of the electro-magnetic spectrum. The pixel array includes n rows of m columns each, starting with the pixel p (1,1) and ending with the pixel p(n, m). When displaying or printing an image, the CPU 12 retrieves the corresponding image file 18 from the memory 16, and operates the monitor 20 or printer 22, as the case may be, as a function of the digital values of the pixels in the image file 18, as is generally known.

According to a feature of the present invention, in an image process, the CPU 12 operates to analyze and process information, for example, the RGB values of the pixels of an image stored in an image file 18, to achieve various objectives, such as, for example, a segregation of illumination and material aspects of a scene depicted in an image of the image file 18 to provide an intrinsic image. The intrinsic image can comprise a material image and/or an illumination image corresponding to the original image, for improved processing in such applications as computer vision. A fundamental observation underlying a basic discovery of the present invention, is that an image comprises two components, material and illumination. All changes in an image are caused by one or the other of these components. Spatio-spectral information is information relevant to contiguous pixels of an image depicted in an image file 18, such as spectral relationships among contiguous pixels, in terms of color bands, for example RGB values of the pixels, and the spatial extent of the pixel spectral characteristics relevant to a characteristic of the image, such as, for example, a single material depicted in the image or illumination flux effecting the image.

What is visible to the human eye upon display of a stored image file 18 by the computer system 10, is the pixel color values caused by the interaction between specular and body reflection properties of material objects in, for example, a scene photographed by the digital camera 14 and illumination flux present at the time the photograph was taken. The illumination flux comprises an ambient illuminant and an incident illuminant. The incident illuminant is light that causes a shadow and is found outside a shadow perimeter. The ambient illuminant is light present on both the bright and dark sides of a shadow, but is more perceptible within the dark region. Pursuant to a feature of the present invention, the computer system 10 can be operated to differentiate between material aspects of the image such as, for example, object edges, and illumination flux through recognition of a spectral shift caused by the interplay between the incident illuminant and the ambient illuminant in the illumination flux.

When one of material and illumination is known in an image, the other can be readily deduced. The spectra for the incident illuminant and the ambient illuminant can be different from one another. Thus, a spectral shift is caused by a shadow, i.e., a decrease of the intensity of the incident illuminant. Pursuant to a feature of the present invention, spectral shift phenomena is captured in spatio-spectral information. The spatio-spectral information includes a spectral ratio: a ratio based upon a difference in color or intensities between two areas of a scene depicted in an image, which may be caused by different materials (an object edge), an illumination change (illumination boundary) or both. In a preferred embodiment of the present invention, the spectral ratio equals Dark/Bright-Dark, where Dark is the color band values (e.g. RGB color values) at a point at an illumination boundary, in a shadowed region of the image, and Bright is the color band values at a point on the illuminated side of the same illumination boundary.

Inasmuch as an illumination boundary is caused by the interplay between the incident illuminant and the ambient illuminant, spectral ratios throughout the image that are associated with illumination change, should be consistently and approximately equal, regardless of the color of the bright side or the material object characteristics of the boundary. A spectral ratio corresponding to an illumination boundary can be designated as a characteristic illuminant ratio for the image.

According to a further feature of the present invention, spatio-spectral operators are generated to facilitate a process for the segregation of illumination and material aspects of a scene depicted in an image file 18. Spatio-spectral operators comprise representations or characteristics of an image that encompass spatio-spectral information usable in the processing of material and illumination aspects of an image. Spatio-spectral operators include, for example tokens.

Pursuant to a feature of the present invention, a token is a connected region of an image wherein the pixels of the region are related to one another in a manner relevant to identification of image features and characteristics such as identification of materials and illumination. The use of tokens recognizes the fact that a particular set of material/illumination/geometric characteristics of an image extends beyond a single pixel, and therefore, while the image processing described herein can be done on a pixel level, tokens expedite a more efficient processing of image properties. The pixels of a token can be related in terms of either homogeneous factors, such as, for example, close correlation of color values among the pixels, or nonhomogeneous factors, such as, for example, differing color values related geometrically in a color space such as RGB space, commonly referred to as a texture. Exemplary embodiments of the present invention provide methods and systems to identify various types of homogeneous or nonhomogeneous tokens for improved processing of image files. The present invention utilizes spatio-spectral information relevant to contiguous pixels of an image depicted in an image file 18 to identify token regions.

According to one exemplary embodiment of the present invention, homogeneous tokens are each classified as either a Type A token, a Type B token or a Type C token. A Type A token is a connected image region comprising contiguous pixels that represent the largest possible region of the image encompassing a single material in the scene. A Type B token is a connected image region comprising contiguous pixels that represent a region of the image encompassing a single material in the scene, though not necessarily the maximal region corresponding to that material. A Type C token comprises a connected image region of similar image properties among the contiguous pixels of the token, for example, similar color and intensity, where similarity is defined with respect to a noise model for the imaging system used to record the image.

Referring now to FIG. 3a, there is shown a flow chart for generating Type C token regions in the image file of FIG. 2a, according to a feature of the present invention. Type C tokens can be readily identified in an image by the CPU 12, utilizing the steps of FIG. 3a. The CPU 12 can then analyze and process the Type C tokens to construct Type B tokens, as will appear.

Prior to execution of the routine of FIG. 3a, the CPU 12 can operate to filter the image depicted in a subject image file 18. The filters may include an image texture filter, to, for example, transform patterns of differing reflectance caused by a textured material into a homogeneous representation that captures the spectral and spatial characteristics of the textured region in the image. Identification of Type B tokens can be difficult in an image texture.

A textured image contains materials with, for example, more than one reflectance function that manifests as a defining characteristic. For example, the defining characteristic can be a pattern of colors within the texture, such that the texture displays a certain distribution of colors in any patch or region selected from anywhere within the textured region of the image. Image texture can be defined as a function of spatial variation in pixel intensities. Image texture patterns are frequently the result of physical or reflective properties of the image surface. Commonly, an image texture is associated with spatial homogeneity and typically includes repeated structures, often with some random variation (e.g., random positions, orientations or colors). Image textures are also often characterized by certain visual properties such as regularity, coarseness, contrast and directionality. An example of image texture is the image of a zebra skin surface as it appears to be spatially homogenous and seems to contain variations of color intensities which form certain repeated patterns. Some image textures can be defined by geometric characteristics, such as stripes or spots.

In many instances, the texture filters may only be required on part of an input image, as much of the image may include homogeneously colored objects. Therefore, prior to application of the texture filters, it is useful to identify and mask off regions of homogeneous color. The texture filters are then only applied to areas where there appear to be textured materials. An example algorithm for identifying textured regions is as follows:

1) Execute a type C tokenization on the N-band color values (e.g. RGB), storing the token results in a region map R, where each pixel in the region map has the token ID of the token to which it belongs (see description of FIG. 3a).

2) Execute a median filter on the region map R (e.g. each pixel P_ij is replaced by the median token ID of a 7×7 box around P_ij). Store the result in R-median.

3) Execute a filter on the original image that calculates the standard deviation of the pixels in a box around each pixel (e.g. 7×7) for each color band. Put the result in S.

4) For each pixel in S, divide the standard deviation calculated for each color band by an estimated noise model value. An example noise model is Sn=A*maxValue +B*pixelValue, where maxValue is the maximum possible color band value, pixelValue is the intensity of a particular band, and A and B are constants experimentally determined for the imaging system (e.g. A32 0.001 and B=0.06 are typical). This step converts the standard deviation into a normalized deviation for each color band. Store the results in Sn.

5) For each pixel in Sn, sum the squares of the normalized deviations for all N color bands, take the square root of the result and divide by the number of bands N to create a deviation value D_ij. Compare the resulting deviation value D_ij to a threshold (e.g. 1.0) assign a 1 to any pixel with a deviation value higher than the threshold, otherwise assign the pixel a 0. Store the results in a texture mask image T.

6) For each pixel in T, if the texture mask value T_ij=1 and the seed size of the token region with the id given in the median region map R-median_ij is less than a threshold (e.g. <4), label the pixel as a textured pixel. Otherwise, label it as a homogeneous pixel. Store the result in the texture mask Tmask.

The output of the above algorithm is a mask, Tmask, which is an array of the size of the original image file 18 (n, m). Pixels (p(n, m)) having a corresponding Tmask location in the array with a 1 value should be treated as part of an image region corresponding to texture materials and pixels with a corresponding Tmask array location having a value of 0 should be treated as part of an image region corresponding to materials of homogeneous color.

The CPU 12 can execute a software module that implements any well known method handling textures, such as, for example, a Laws filter bank, or wavelets (see, for example, Randen, T.[Trygve], Husøy, J. H.[John Håkon], Filtering for Texture Classification: A Comparative Study, PAMI(21), No. 4, April 1999, pp. 291-310). See also: Are Filter Banks Necessary? Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, June 2003, volume 2, pages 691-698.

In one exemplary embodiment of the present invention, Type C tokens are generated using an intensity histogram representation for each pixel of an image (or, preferably, for pixels of the image in regions identified as comprising a texture). The Type C tokens generated using intensity histograms are then used to generate Type B tokens. An intensity histogram representation for each pixel in a texture filter output image can be produced using the following algorithm:

• • A) Loop through all memory addresses/indexes in the Tmask array (n, m);
  • B) At each memory address/index of the Tmask array determine whether the value at each location is 1 or 0;
  • C) If a memory address/index of the Tmask has a 1 value indicating texture:
    • a. Open the corresponding memory address/index in the corresponding image file 18 p(n, m) and access that memory address/index;
    • b. Open that memory address/index for writing data;
    • c. Initialize N 8-bin histograms to zero, one for each color band;
    • d. For each pixel q within a neighborhood of p (for example, an 11×11 pixel array):
    • For each of the N color values Cn of q (e.g. RGB wherein each color value is set between 0-255);
      • (i) If N color value is between 0 and 31, then increment the first bin in 8-bin histogram of color N;
      • (ii) If N color value is between 32 and 64, then increment the second bin in 8-bin histogram of color N;
      • (iii) If N color value is between 65 and 97, then increment the third bin in 8-bin histogram of color N;
      • (iv) If N color value is between 98 and 130, then increment the fourth bin in 8-bin histogram of color N;
      • (v) If N color value is between 131 and 163, then increment the fifth bin in 8-bin histogram of color N;
      • (vi) If N color value is between 164 and 196, then increment the sixth bin in 8-bin histogram of color N;
      • (vii) If N color value is between 197 and 229, then increment the seventh bin in 8-bin histogram of color N;
      • (viii) If N color value is between 230 and 255, then increment the eighth bin in 8-bin histogram of color N;
    • e. Concatenate the N 8-bin histogram values together into a single intensity histogram comprising an 8×N element vector;
    • f. Assign the 8×N element vector to the corresponding pixel p in the image file array p(n, m);
    • g. If current or next memory address/index is not last/Tmask (n, m), then move on to the next memory address/index in Tmask (n, m) array and execute step B;
    • h. If current memory address/index is last/equal to Tmask (n, m), then cease looping through Tmask (n, m) array.
  • D) If a memory address/index has a 0 value:
    • i. If current address/index is not a last element in Tmask (n, m) array, then move on to the next memory address/index in Tmask (n, m) array and execute step B;
    • ii. If current memory address/index location is a last element in Tmask (n, m) array, then cease looping through Tmask (n, m) array.

After the transformation for each pixel in the textured region of the image from a set of color bands, for example RGB, to an intensity histogram vector comprising a set of filter outputs, for example the 8×N concatenation, the image is treated exactly as the original color band image with respect to identifying type C tokens with the 8×N element intensity histogram vector being used in place of the RGB color band values.

FIG. 2b shows an 11×11 pixel array at the upper left hand corner of an image file 18. The 11×11 pixel array of FIG. 2b is used to produce an intensity histogram representation for pixel p(6,6) at the center Z of the array, according to the exemplary algorithm described above. In our example, it is assumed that the 11×11 array is within a checkerboard pattern within the image depicted in the image file 18. The checkerboard pattern comprises a pattern of alternating red, green and blue boxes as shown. Thus, pixel p(6,6) of our example will have a 1 value in the corresponding Tmask generated by the CPU 12.

Pursuant to the exemplary algorithm, three 8 bin histograms are initialized by the CPU 12 (see FIG. 2c), one for each of the red, green and blue color values of the subject pixel p(6,6). The 11×11 pixel array is then used to determine the values within the bins of the three histograms. In our example, the upper left hand section of the 11×11 array is a red box of the checkerboard pattern, and each pixel within that section (p(1,1) to p(5,5)) has an RGB value of (123,0,0). The upper right hand section is green, thus the pixels within the section p(1,7) to p(5,11)) each have an RGB value of (0,123, 0). The lower left hand section is a blue box, with each of the pixels p(7,1) to p(11,5) having RGB values of (0,0,123). Finally the lower right hand section of our example is again a red section, with each of the pixels p(7,7) to p(11,11) having RGB values of (123,0,0), in the repeating red, green and blue checkerboard pattern.

As shown in FIG. 2b, the one (1) pixel thick cross-shaped border lines between red quadrant (p(1,1) to p(5,5)), green quadrant (p(1,7) to p(5,11)), blue quadrant (p(7,1) to p(11,5)) and red quadrant (p(7,7) to p(11,11)) or any pixel between p(6,1) to p(6,m) and any pixel between p(n,6) to p(11,6) may have an RGB value of (123,123,123). Pixel p(6,6) at the center Z may also have the RGB value of (123,123,123).

Next, an 11×11 pixel box 301, with pixel Z at its center, is used to determine the value of pixel Z via an histogram representation of pixel Z's RGB color band values using the steps exemplarily described below. As shown in FIG. 2b, p(1,1) has RGB color value of (123,0,0) indicating some presence of red color and absence of green color and blue color. Since the red value of p(1,1) is 123, the corresponding bin for that number value in red 8 bin histogram, as shown in FIG. 2c, would be incremented i.e. fourth bin within the red 8-bin histogram would be incremented. This process would repeat itself until p(n,m) in pixel box 301 is reached. By this time, three 8-bin histograms, as shown in FIG. 2c, may be filled with different values. These values may then be concatenated together into a single 8×3 element vector and assigned to pixel Z in 11×11 pixel array depicted in the image file 18 of FIG. 2b.

Consequently, according to the exemplary algorithm described above and a checkerboard pattern within 11×11 pixel box 301 shown within image 18 depicted in FIG. 2b, the 3 8-bin histogram would be filled as follows:

In red 8-bin histogram, the fourth bin would have a value of 71(the amount of repetitions of red values between 98 and 130 in 25 pixels of 2 red quadrants plus the amount of repetitions of red values between 98 and 130 in 11 pixels between p(6,1) to p(6,m) plus the amount of repetitions of red values between 98 and 130 in 10 pixels between p(n,6) to p(11,6) resulting in 25+25+11+10=71). In a similar calculation, the first bin would have a value of 50 to indicate all of the pixels with a zero value for red. All the other bins in red 8-bin histogram would have the values of 0. Thus, the red 8-bin histogram would be [50, 0, 0, 71, 0, 0, 0, 0].

In green 8-bin histogram, the fourth bin would have a value of 46 (the amount of repetitions of green values between 98 and 130 in 25 pixels of the green quadrant plus the amount of repetitions of green values between 98 and 130 in 11 pixels between p(6,1) to p(6,m) plus the amount of repetitions of green values between 98 and 130 in 10 pixels between p(n,6) to p(11,6) resulting in 25+11+10=46). The first bin has a value of 75 (for zero values of green). All the other bins in green 8-bin histogram would have the values of 0. Thus, the green 8-bin histogram would be [75, 0, 0, 46, 0, 0, 0, 0].

In blue 8-bin histogram, the fourth bin would have a value of 46 (the amount of repetitions of blue values between 98 and 130 in 25 pixels of the blue quadrant plus the amount of repetitions of blue values between 98 and 130 in 11 pixels between p(6,1) to p(6,m) plus the amount of repetitions of blue values between 98 and 130 in 10 pixels between p(n,6) to p(11,6) resulting in 25+11+10=46). The first bin, again, has a value of 75. All the other bins in blue 8-bin histogram would have the values of 0. Thus, the blue 8-bin histogram would be [75, 0, 0, 46, 0, 0, 0, 0].

As a result of the above described algorithm, each 8-bin histogram, three in our example, would be concatenated together into a single 8×3 element vector. Accordingly, this vector would be [50, 0, 0, 71, 0, 0, 0, 0, 75, 0, 0, 46, 0, 0, 0, 0, 75, 0, 0, 46, 0, 0, 0, 0].

Next, the above described vector would be assigned at p(6,6) at the center Z of the 11×11 array and therefore replace the RGB value of pixel p(6,6), which originally was (123, 123, 123), with an intensity histogram set forth as [50, 0, 0, 71, 0, 0, 0, 0, 75, 0, 0,46, 0, 0, 0, 0, 75, 0, 0, 46, 0, 0, 0, 0].

A 1^st order uniform, homogeneous Type C token comprises a single robust color measurement among contiguous pixels of the image. At the start of the identification routine of FIG. 3a, the CPU 12 sets up a region map in memory. In step 100, the CPU 12 clears the region map and assigns a region ID, which is initially set at 1. An iteration for the routine, corresponding to a pixel number, is set at i=0, and a number for an N×N pixel array, for use as a seed to determine the token, is set an initial value, N=N_start. N_start can be any integer>0, for example it can be set at set at 11 or 15 pixels.

At step 102, a seed test is begun. The CPU 12 selects a first pixel, i=(1, 1) for example (see FIG. 2a), the pixel at the upper left corner of a first N×N sample of the image file 18. The pixel is then tested in decision block 104 to determine if the selected pixel is part of a good seed. The test can comprise a comparison of the color value of the selected pixel to the color values of a preselected number of its neighboring pixels as the seed, for example, the N×N array. The color values comparison can be with respect to multiple color band values (RGB in our example) of the pixel or the filter output intensity histogram representation of the pixel, in the event the image was filtered for texture regions, as described above. If the comparison does not result in approximately equal values (for example, within the noise levels of the recording device for RGB values) for the pixels in the seed, the CPU 12 increments the value of i (step 106), for example, i=(1, 2), for a next N×N seed sample, and then tests to determine if i=i_max (decision block 108).

If the pixel value is at i_max, a value selected as a threshold for deciding to reduce the seed size for improved results, the seed size, N, is reduced (step 110), for example, from N=15 to N=12. In an exemplary embodiment of the present invention, i_max can be set at i=(n, m). In this manner, the routine of FIG. 5a parses the entire image at a first value of N before repeating the routine for a reduced value of N.

After reduction of the seed size, the routine returns to step 102, and continues to test for token seeds. An N_stop value (for example, N=2) is also checked in step 110 to determine if the analysis is complete. If the value of N is at N_stop, the CPU 12 has completed a survey of the image pixel arrays and exits the routine.

If the value of i is less than i_max, and N is greater than N_stop, the routine returns to step 102, and continues to test for token seeds.

When a good seed (an N×N array with approximately equal pixel values) is found (block 104), the token is grown from the seed. In step 112, the CPU 12 pushes the pixels from the seed onto a queue. All of the pixels in the queue are marked with the current region ID in the region map. The CPU 12 then inquires as to whether the queue is empty (decision block 114). If the queue is not empty, the routine proceeds to step 116.

In step 116, the CPU 12 pops the front pixel off the queue and proceeds to step 118. In step 118, the CPU 12 marks “good” neighbors around the subject pixel, that is neighbors approximately equal in color value to the subject pixel, with the current region ID. All of the marked good neighbors are placed in the region map and also pushed onto the queue. The CPU 12 then returns to the decision block 114. The routine of steps 114, 116, 118 is repeated until the queue is empty. At that time, all of the pixels forming a token in the current region will have been identified and marked in the region map as a Type C token. In the event the pixels comprise intensity histogram representations, the token can be marked as Type C_T.

When the queue is empty, the CPU 12 proceeds to step 120. At step 120, the CPU 12 increments the region ID for use with identification of a next token. The CPU 12 then returns to step 106 to repeat the routine in respect of the new current token region.

Upon arrival at N=N_stop, step 110 of the flow chart of FIG. 3a, or completion of a region map that coincides with the image, the routine will have completed the token building task. FIG. 3b is an original image used as an example in the identification of tokens. The image shows areas of the color blue and the blue in shadow, and of the color teal and the teal in shadow. FIG. 3c shows token regions corresponding to the region map, for example, as identified through execution of the routine of FIG. 3a (Type C tokens), in respect to the image of FIG. 3b. The token regions are color coded to illustrate the token makeup of the image of FIG. 3b, including penumbra regions between the full color blue and teal areas of the image and the shadow of the colored areas.

Prior to commencing any process to generate Type B tokens from the identified Type C tokens, the CPU 12 tests each identified Type C token to make certain that each Type C token encompasses a single material. While each Type C token comprises a region of the image having a single robust color measurement among contiguous pixels of the image, the token may grow across material boundaries.

Typically, different materials connect together in one Type C token via a neck region often located on shadow boundaries or in areas with varying illumination crossing different materials with similar hue but different intensities. A neck pixel can be identified by examining characteristics of adjacent pixels. When a pixel has two contiguous pixels on opposite sides that are not within the corresponding token, and two contiguous pixels on opposite sides that are within the corresponding token, the pixel is defined as a neck pixel.

FIG. 4 shows a flow chart for a neck test for Type C tokens. In step 122, the CPU 12 examines each pixel of an identified token to determine whether any of the pixels under examination forms a neck. The routine of FIG. 4 can be executed as a subroutine directly after a particular token is identified during execution of the routine of FIG. 3a. All pixels identified as a neck are marked as “ungrowable.” In decision block 124, the CPU 12 determines if any of the pixels were marked.

If no, the CPU 12 exits the routine of FIG. 4 and returns to the routine of FIG. 3a (step 126).

If yes, the CPU 12 proceeds to step 128 and operates to regrow the token from a seed location selected from among the unmarked pixels of the current token, as per the routine of FIG. 3a, without changing the counts for seed size and region ID. During the regrowth process, the CPU 12 does not include any pixel previously marked as ungrowable. After the token is regrown, the previously marked pixels are unmarked so that other tokens may grow into them.

Subsequent to the regrowth of the token without the previously marked pixels, the CPU 12 returns to step 122 to test the newly regrown token.

Neck testing identifies Type C tokens that cross material boundaries, and regrows the identified tokens to provide single material Type C tokens suitable for use in creating Type B tokens. FIG. 3d shows Type B tokens generated from the Type C tokens of FIG. 3c, according to a feature of the present invention. The present invention provides several exemplary techniques of pixel characteristic analysis for constructing Type B tokens from Type C tokens. One exemplary technique involves arbitrary boundary removal. The arbitrary boundary removal technique can be applied to Type C tokens whether they were generated using N color band values (RGB in our example) of the pixel or the filter output representation of the pixel, in the event the image was filtered. Actual boundaries of any particular Type C token will be a function of the seed location used to generate the token, and are thus, to some extent arbitrary. There are typically many potential seed locations for each particular token, with each potential seed location generating a token with slightly different boundaries and spatial extent because of differences among the color values of the pixels of the various seeds, within the noise ranges of the recording equipment.

FIG. 5 is a flow chart for constructing Type B tokens via an arbitrary boundary removal technique, according to a feature of the present invention. In step 200, the CPU 12 is provided with a set (T_c) of Type C tokens generated with a seed size (S) via the routine of FIG. 3a, with neck removal via the routine of FIG. 4. The seed size S=S_max, for example, S=4 pixels. In step 202, for each Type C token, t_c in the set T_c the CPU 12 selects a number (for example 50) of potential seeds s₁ to s_n. In our example, each selected seed will be a 4×4 pixel array from within the token region, the pixels of the array being of approximately equal values (within the noise levels of the recording device).

In step 204, the CPU 12 grows a new Type C token, utilizing the routines of FIGS. 3a and 4, from each seed location, s₁ to s_n of each token t_c in the set T_c. The newly grown tokens for each token t_c are designated as tokens r_c1 to r_cn. The newly grown tokens r_c1 to r_cn for each token t_c generally overlap the original Type C token t_c, as well as one another.

In step 206, the CPU 12 operates to merge the newly generated tokens r_c1 to r_cn of each token t_c, respectively. The result is a new token R_t corresponding to each original token t_c in the set T_c. Each new token R_t encompasses all of the regions of the respective overlapping tokens r_c1 to r_cn generated from the corresponding original token t_c. The unions of the regions comprising the respective merged new tokens R_t are each a more extensive token than the original Type C tokens of the set. The resulting merged new tokens R_t result in regions of the image file 18, each of a much broader range of variation between the pixels of the respective token R_t than the original Type C token, yet the range of variation among the constituent pixels will still be relatively smooth. R_t is defined as a limited form of Type B token, Type B_ab1, to indicate a token generated by the first stage (steps 200-206) of the arbitrary boundary removal technique according to a feature of the present invention.

In step 208, the CPU 12 stores each of the Type B_ab1 tokens generated in steps 202-206 from the set of tokens T_c, and proceeds to step 210. Type B_ab1 tokens generated via execution of steps 202-206 may overlap significantly. In step 210, the CPU 12 operates to merge the R_t tokens stored in step 208 that overlap each other by a certain percentage of their respective sizes. For example, a 30% overlap is generally sufficient to provide few, if any, false positive merges that combine regions containing different materials. The new set of merged tokens still may have overlapping tokens, for example, previously overlapping tokens that had a less than 30% overlap. After all merges are complete, the CPU 12 proceeds to step 212.

In step 212, the CPU 12 identifies all pixels that are in more than one token (that is in an overlapping portion of two or more tokens). Each identified pixel is assigned to the token occupying the largest region of the image. Thus, all overlapping tokens are modified to eliminate all overlaps.

In step 214, the CPU 12 (as the Type C tokenization block 35 or the operators block 28) stores the final set of merged and modified tokens, now designated as Type B_ab2 tokens, and then exits the routine. As noted above, the Type B_ab2 tokens were generated from Type C tokens whether the Type C tokens were generated using N color band values (RGB in our example) of the pixel or the filter output representation of the pixel, in the event the image was filtered.

A second exemplary technique according to the present invention, for using Type C tokens to create Type B tokens, is adjacent planar token merging. The adjacent planar token merging can be implemented when an image depicts areas of uniform color, that is for non-textured regions of an image. Initially, a token graph is used to identify tokens that are near to one another. FIG. 6 shows a flow chart for creating a token graph, containing token map information, according to a feature of the present invention. Each token t_c in the set of Type C tokens T_c, generated through execution of the routines of FIGS. 3a and 4, is evaluated in terms of a maximum distance D_max between tokens defining a neighboring pair of tokens, t_c, t_n, of the set T_c, a minimum number of token perimeter pixels, P_min, in each token of the neighboring pair of tokens, and a minimum fraction of perimeter pixels, F_min, of each token of a neighboring pair of tokens, required to be within D_max.

In step 300, the CPU 12 selects a Type C token t_c in the set of Type C tokens T_c, and identifies the pixels of the selected token t_c forming the perimeter of the token. In a decision block 302, the CPU 12 determines whether the number of perimeter pixels is less than P_min, for example 10pixels.

If yes, the CPU 12 proceeds to decision block 304 to determine whether there are any remaining tokens t_c in the set of Type C tokens T_c. If yes, the CPU 12 returns to step 300, if no, the CPU 12 exits the routine 306.

If no, the CPU 12 proceeds to step 308. In step 308, the CPU 12 generates a bounding box used as a mask to surround the selected token t_c. The bounding box is dimensioned to be at least D_max larger than the selected token t_c in all directions. A known distance transform (for example, as described in P. Felzenszwalb and D. Huttenlocher, Distance Transforms of Sampled Functions, Cornell Computing and Information Science Technical Report TR2004-1963, September 2004), is executed to find the distance from each perimeter pixel of the selected token t_c to all the pixels in the surrounding bounding box. The output of the distance transform comprises two maps, each of the same size as the bounding box, a distance map and a closest pixel map. The distance map includes the Euclidean distance from each pixel of the bounding box to the nearest perimeter pixel of the selected token t_c. The closest pixel map identifies, for each pixel in the distance map, which perimeter pixel is the closest to it.

In step 310, the CPU 12 scans the distance map generated in step 308 to identify tokens corresponding to pixels of the bounding box (from the region map generated via the routine of FIG. 3a), to identify a token from among all tokens represented by pixels in the bounding box, that has a number N_cn of pixels within the distance D_max, wherein N_cn is greater than P_min, and greater than F_min * perimeter pixels of the respective token and the average distance between the respective token and t_c is the lowest of the tokens corresponding to the pixels in the bounding box. If these conditions are satisfied, the respective token is designated t_n of a possible token pair t_c, t_n, and a link L_cn is marked active.

In step 312, the CPU 12 checks to determine whether a reciprocal link L_cn is also marked active, and when it is marked active, the CPU 12 marks and stores in the token graph, an indication that the token pair t_c, t_n is a neighboring token pair. The reciprocal link refers to the link status in the evaluation of the token designated as t_n in the current evaluation. If that token has yet to be evaluated, the pair is not designated as a neighboring token pair until the link L_cn is verified as active in the subsequent evaluation of the token t_n. The CPU 12 then returns to decision block 304 to determine whether there are any further tokens in the set T_c.

Upon completion of the token graph, the CPU 12 utilizes token pair information stored in the graph in the execution of the routine of FIG. 7. FIG. 7 shows a flow chart for constructing Type B tokens via the adjacent planar token merging technique, according to a feature of the present invention. In the adjacent planer merging technique, pairs of tokens are examined to determine wether there is a smooth and coherent change in color values, in a two dimensional measure, between the tokens of the pair. The color change is examined in terms of a planar representation of each channel of the color, for example the RGB components of the pixels according to the exemplary embodiments of the present invention. A smooth change is defined as the condition when a set of planes (one plane per color component) is a good fit for the pixel values of two neighboring tokens. In summary, neighboring tokens are considered the same material and a Type B token when the color change in a two-dimensional sense is approximately planar.

In step 320, the CPU 12 selects a token pair t_c, t_n from the token graph. In decision block 322, the CPU 12 determines whether the mean color in token t_c is significantly different from the mean color in the token t_c. The difference can be a function of a z-score, a known statistical measurement (see, for example, Abdi, H. (2007), Z-scores, in N. J. Salkind (Ed.), Encyclopedia of Measurement and Statistics, Thousand Oaks, Calif.: Sage), for example, a z-score greater than 3.0.

If the mean colors of the token pair are different, the CPU 12 proceeds to decision block 324 to determine whether there are any additional token pairs in the token graph. If yes, the CPU 12 returns to step 320. If no, the CPU 12 exits the routine (step 326).

If the mean colors are within the z-score parameter, the CPU 12 proceeds to step 328. In step 328, the CPU 12 performs a mathematical operation such as, for example, a least median of squares regression (see, for example, Peter J. Rousseeuw, Least Median of Squares Regression, Journal of the American Statistical Association, Vol. 79, No. 388 (December, 1984), pp. 871-880) to fit a plane to each color channel of the pixels (in our example RGB) of the token pair t_c, t_n, as a function of row n and column m (see FIG. 2), the planes being defined by the equations:
R=X_Rn+Y_Rm+Z_R G=X_Gn+Y_Gm+Z_G B=X_Bn+Y_Bm+Z_B
wherein parameter values X, Y and C are determined by the least median of squares regression operation of the CPU 12.

Upon completion of the plane fitting operation, the CPU 12 proceeds to step 330. In step 330, the CPU 12 examines each pixel of each of the tokens of the token pair t_c, t_n to calculate the z-score between each pixel of the tokens and the planar fit expressed by the equation of the least median of squares regression operation. When at least a threshold percentage of the pixels of each token of the pair (for example, 80%), are within a maximum z-score (for example, 0.75), then the neighboring token pair is marked in the token graph as indicating the same material in the image. After completion of step 330, the CPU 12 returns to decision block 324.

Upon exiting the routine of FIG. 7, the CPU 12 examines the token graph for all token pairs indicating the same material. The CPU 12 can achieve the examination through performance of a known technique such as, for example, a union find algorithm. (See, for example, Zvi Galil and Giuseppe F. Italiano. Data structures and algorithms for disjoint set union problems, ACM Computing Surveys, Volume 23, Issue 3 (September 1991), pages 319-344). As a simple example, assume a set of seven Type C tokens T₁, T₂, T₃, T₄, T₅, T₆, T₇. Assume that the result of the execution of FIG. 9, (performance of the adjacent planar analysis), indicates that tokens T₁ and T₂ are marked as the same material, and tokens T₁ and T₃ are also marked as the same material. Moreover, the results further indicate that tokens T₄ and T₅ are marked as the same material, and tokens T₅ and T₆ are also marked as the same material. The result of execution of the union find algorithm would therefore indicate that tokens {T₁, T₂, T₃} form a first group within the image consisting of a single material, tokens {T₄, T₅, T₆} form a second group within the image consisting of a single material, and token {T₇} forms a third group within the image consisting of a single material. The groups {T₁, T₂, T₃}, {T₄, T₅, T₆} and {T₇} form three Type B tokens.

A third exemplary technique according to the present invention, for using Type C tokens to create Type B tokens, is a local token analysis. A local token approach generates Type C tokens using a window analysis of a scene depicted in an image file 18. Such tokens are designated as Type C_w tokens. FIG. 8 is a flow chart for generating Type C_w tokens via the local token analysis technique, according to a feature of the present invention.

In step 400, the CPU 12 places a window of fixed size, for example, a 33×33 pixel array mask, over a preselected series of scan positions over the image. The window can be a shape other than a square. The scan positions are offset from one another by a fixed amount, for example ½ window size, and are arranged, in total, to fully cover the image. The window area of pixels at each scan position generates a Type C_w token, though not every pixel within the window at the respective scan position is in the Type C_w token generated at the respective scan position.

At each scan position (step 402), the CPU 12 operates, as a function of the pixels within the window, to fit each of a set of planes, one corresponding to the intensity of each color channel (for example, RGB), and an RGB line in RGB space, characterized by a start point I₀ and an end point I₁ of the colors within the window. The planar fit provides a spatial representation of the pixel intensity within the window, and the line fit provides a spectral representation of the pixels within the window.

For the planar fit, the planes are defined by the equations:
R=X_Rn+Y_Rm+Z_R G=X_Gn+Y_Gm+Z_G B=X_Bn+Y_Bm+Z_B
wherein parameter values X, Y and C are determined by CPU 12 by executing a mathematical operation such as the least median of squares regression discussed above, a least-squares estimator, such as singular value decomposition, or a robust estimator such as RANSAC (see, for example, M. A. Fischler, R. C. Bolles. Random Sample Consensus: A Paradigm for Model Fitting with Applications to Image Analysis and Automated Cartography. Comm. of the ACM, Vol 24, pp 381-395, 1981).

For the RGB line fit, the line is defined by: I(r,g,b)=I₀(r,g,b)+t(I₁(r,g,b)−I₀(r,g,b)) wherein the parameter t has a value between 0 and 1, and can be determined by the CPU 12 utilizing any of the mathematical techniques used to find the planar fit.

At each scan position, after completion of step 402, the CPU 12 operates in step 404 to examine each pixel in the window in respect of each of the planar fit representation and RGB line representation corresponding to the respective window scan position. For each pixel, the CPU 12 determines an error factor for the pixel relative to each of the established planes and RGB line. The error factor is related to the absolute distance of the pixel to its projection on either from either the planar fit or the RGB line fit. The error factor can be a function of the noise present in the recording equipment or be a percentage of the maximum RGB value within the window, for example 1%. Any pixel distance within the error factor relative to either the spatial planar fit or the spectral line fit is labeled an inlier for the Type C_w token being generated at the respective scan position. The CPU 12 also records for the Type C_w token being generated at the respective scan position, a list of all inlier pixels.

At each scan position, after completion of step 404, the CPU 12 operates in step 406 to assign a membership value to each inlier pixel in the window. The membership value can be based upon the distance of the inlier pixel from either the planar fit or the RGB line fit. In one exemplary embodiment of the present invention, the membership value is the inverse of the distance used to determine inlier status for the pixel. In a second exemplary embodiment, a zero-centered Gaussian distribution with a standard deviation is executed to calculate membership values for the inlier pixels.

After all of the scan positions are processed to generate the Type C_w tokens, one per scan position, the CPU 12 operates to compile and store a token data list (step 408). The token data list contains two lists. A first list lists all of the pixels in the image file 18, and for each pixel, an indication of each Type C_w token to which it labeled as an inlier pixel, and the corresponding membership value. A second list lists all of the generated Type C_w tokens, and for each token an indication of the inlier pixels of the respective token, and the corresponding membership value. After compiling and storing the token data list, the CPU 12 exits the routine (step 410).

FIG. 9 is a flow chart for constructing Type B tokens from the Type C_w tokens generated via the local token analysis technique, according to a feature of the present invention. In step 420, the CPU 12 calculates a similarity of parameters of the spatial planer dimensions and spectral RGB lines of adjacent or overlapping Type C_w tokens generated through execution of the routine of FIG. 108. Overlapping and adjacent Type C_w tokens can be defined as tokens corresponding to scan positions that overlap or are contiguous. A similarity threshold can be set as a percentage of difference between each of the spatial planer dimensions and spectral RGB lines of two overlapping or adjacent Type C_w tokens being compared. The percentage can be a function of the noise of, for example, the camera 14 used to record the scene of the image file 18. All overlapping or adjacent Type C_w token pairs having a calculated similarity within the similarity threshold are placed on a list.

In step 422, the CPU 12 sorts the list of overlapping or adjacent Type C_w token pairs having a calculated similarity within the similarity threshold, in the order of most similar to least similar pairs. In step 424, the CPU 12 merges similar token pairs, in the order of the sort, and labeling pairs as per degree of similarity. Each merged token pair will be considered a Type_B token. In step 426, the CPU 12 stores the list of Type_B tokens, and exits the routine.

In a further exemplary embodiment of the present invention, the CPU 12 compiles lists of Type B tokens separately generated through each of and/or a combination of one or more of the arbitrary boundary removal, adjacent planar token merging, and local token analysis techniques. The determination of the combination of techniques used depends in part on whether a particular region of the image was filtered because of texturing of the image. Since each Type B token generated through the described techniques likely represents a single material under varying illumination conditions, merging sufficiently overlapping Type B tokens generated through the use of varying and different techniques, provides a resulting, merged Type B token that represents a more extensive area of the image comprising a single material, and approaches the extent of a Type A token.

Sufficiently overlapping can be defined by satisfaction of certain pixel characteristic criteria, such as, for example:

• A) The two Type B tokens have at least n of the original Type C tokens in common, for example, n=1
• B) The two Type B tokens have at least n pixels in common, for example, n=20
• C) The two Type B tokens have at least n % overlap, that is at least n % of the pixels in a first one of the two Type B tokens are also found in the second one of the two Type B tokens or vice versa, wherein, for example n %=10%.
• D) The percentage of pixels in a smaller one of the two Type B tokens, also found in the larger one of the two Type B tokens is above a preselected threshold, for example 15%.
• E) A preselected combination of criteria A-D.

Merging of two sufficiently overlapping Type B tokens can be accomplished via a mathematical operation such as execution of the union find algorithm discussed above. In the case of two overlapping Type B tokens that do not satisfy the above discussed criteria, the overlapping pixels of the two tokens can be assigned to the larger one of the two Type B tokens.

As a result of execution by the CPU 12 of the token generation and merging techniques according to features of the present invention, an image can be accurately segmented into tokens representing discrete materials depicted in the scene (Type B tokens) and tokens representing regions of robust similar color (Type C tokens), thus providing a basis for computational efficiencies, as the token representations capture spatio-spectral information of a significant number of constituent pixels.

These token operators provide an excellent representation of images that include large surface areas of a single material, such as are often depicted in images including man-made objects. However, in many natural scenes there are often large areas of highly textured regions, such as sand, grass, stones, foliage, and so on. As noted above, identification of Type B tokens using Type C tokens, can be difficult in an image texture. According to a further feature of the present invention, a texton histogram operator provides a mechanism for capturing statistically uniform spatial variations of textured regions in a manner that is useful in an image segregation operation.

Thus, according to this feature of the present invention, rather than generating Type C tokens in textured regions of an image, from intensity histograms, for use in identifying Type B tokens, as described above, texture tokens are generated as a species of Type B tokens. In an exemplary embodiment of the texton histogram operator, the CPU 12 converts each pixel of the image (or pixels of those regions of an image identified as comprising a texture) from the recorded color band representation of the respective image file 18, such as, for example, RGB color band values, to a two band representation wherein the two bands comprise a texton label and a texton histogram label. The two band representations for the pixels are then used to identify texture tokens, as will appear.

A texton label for each pixel is generated through execution of a clustering process. A texture can be characterized by a texture primitive (for example, in a grass texture, a single blade of grass), and the spatial distribution of the primitive. A texton analysis is an analytical method for characterizing a texture primitive, for example via a clustering algorithm. Clustering is a process for locating centers of natural groups or clusters in data. In an exemplary embodiment of the present invention, the data comprises pixel patches selected from among the pixels of an image being segregated into material and illumination components. For example, 3×3 pixel patches are clustered into K different groups, with each group being assigned a designating number (1,2,3, . . . K). The texton label for each pixel of the 3×3 array is the group number of the group to which the respective patch was assigned during the clustering process.

To expedite execution of a clustering algorithm, random samples of 3×3 patches can be selected throughout the image, or region of the image identified as comprising a texture, for processing in a clustering algorithm. After execution of the clustering algorithm by the CPU 12, each 3×3 patch of the image is assigned the texton label of the closest one of the K group centers identified in the clustering process, as executed in respect of the selected random samples.

To advantage, prior to execution of a clustering algorithm, the pixels of the image are subject to an image intensity normalization. In a clustering process utilizing an intensity-based distance matrix, dark areas of an image may be placed in a single group, resulting in an under representation of groups for shadowed areas of a textured region of an image. A normalization of the image provides a more accurate texton representation for texture regions under varying illumination. A normalized intensity for a pixel can be expressed by:
i_norm(n,m)=log(i(n,m)/i_b(n,m)),
where i_norm(n,m) is the normalized intensity for a pixel p(n,m), i(n,m) is the intensity for the pixel p(n,m), as recorded in the image file 18, and i_b(n,m) is a blurred or low passed filtered version of the pixel p(n,m). For example, a 10 pixel blur radius can be used in any standard blurring function.

Clustering can be executed according to any known clustering algorithm, such as, for example, K means clustering where there are K clusters or groups S_i, i=1,2, . . . K, and μ_i is the mean point or center point of all the data points x_j ε S_i. In our example, each x_j comprises a selected 3×3 pixel patch arranged as a 9×1 vector of the nine pixels in the patch (27 elements total, including the RGB values of each of the nine pixels of the vector). As noted above, each mean point μ_i is assigned a texton label, 1, 2, 3. . . K, that becomes the texton label for any pixel of a 3×3 patch clustered into the group for which the respective mean point is the center.

According to an exemplary embodiment of the present invention, the CPU 12 executes the algorithm by initially partitioning the selected 9×1 vectors, representing 3×3 pixel patches of the image, into K initial groups S_i. The CPU 12 then calculates a center point μ_i, for each group S_i utilizing an intensity-based distance matrix. After determining a center point μ_i, for each group S_i, the CPU 12 associates each 9×1 vector to the closest center point μ_i, changing groups if necessary. Then the CPU 12 recalculates the center points μ_i. The CPU 12 executes iterations of the steps of associating each 9×1 vector to the closest center point μ_i, and recalculating the center points μ_i, until convergence. Convergence is when there is no need to change the group for any of the 9×1 vectors. At that point, the CPU 12 assigns the group number for the respective center point μ_i, as the texton label for the pixels of each vector in that group.

As noted above, pixels of 3×3 patches not selected as samples for clustering are assigned the texton label of the closest one of the K group centers μ_i, identified in the clustering process, as executed in respect of the selected random samples. A texton label map is stored by the CPU 12, and is coextensive with the pixel array of FIG. 2a. In the texton label map, for each pixel location, there is an indication of the respective texton label.

Upon completion of the texton label assignment for pixels of the image, the CPU 12 operates to generate a texton histogram for each pixel to provide a representation of the spatial variation of texton representations within a textured region of the image. To that end, the CPU 12 accesses the texton label map. At each pixel location within the texton label map, a pixel patch of, for example, 21×21 pixels, is set up around the current location. The 21×21 patch size is far greater than the 3×3 patch sized used to generate the texton representations, so as to capture the spatial variations of the texture. A texton histogram is then generated for the pixel location at the center of the 21×21 patch, in a similar manner as the intensity histogram described above. However, rather than bins based upon color band values, in the texton histogram, there is a bin for each texton label value, 1,2,3 . . . K. The count for each bin corresponds to the number of pixels in the 21×21 patch having the texton label value for the respective bin.

When a texton histogram is generated for each pixel of the texton label map, the CPU 12 executes a second clustering step. In the second clustering step, the texton histograms are clustered using spectral clustering. Spectral clustering techniques use a spectrum of a similarity matrix of data of interest, (in our example, the texton histograms) to reduce the dimensionality for clustering in fewer dimensions. A similarity matrix for a given set of data points A can be defined as a matrix S where S_ij represents a measure of the similarity between points i, j ε A. In our example, eigenvectors of the Laplacian are clustered using a mean shift. The distance metric is a chi-squared distance of the histograms.

A texton histogram label (1,2. . .) is assigned to each cluster group defined by the clustering procedure. For each pixel of the texton label map, the texton histogram label for the cluster group corresponding to a texton histogram that is nearest the texton histogram for the respective pixel, is assigned to that pixel. Distance is defined as the chi-squared histogram distance. Upon completion of the assignment of a texton histogram label to each pixel, each pixel is now represented by a two band, texton label, texton histogram label representation.

It can be assumed that a region of an image wherein contiguous pixels within the region all have the same two band, texton label, texton histogram label representation, comprises a region of the same mean material of a texture depicted in the image. Such a region can be referred to as a texture token, a species of a Type B token.

In the preceding specification, the invention has been described with reference to specific exemplary embodiments and examples thereof. It will, however, be evident that various modifications and changes may be made thereto without departing from the broader spirit and scope of the invention as set forth in the claims that follow. The specification and drawings are accordingly to be regarded in an illustrative manner rather than a restrictive sense.

Claims

1. An automated, computerized method for processing an image, comprising the steps of:

a computer executing the following steps:

converting a color band representation of each pixel in the image to a texton representation, the converting to a texton representation being carried out by generating a two band, texton label, texton histogram label representation for each pixel in the image, wherein the texton label is generated by a first clustering process and represents a texture primitive of the texture in the image and the texton histogram label is generated by a second clustering process and represents a spatial distribution of the texture primitive within the image; and

utilizing the texton representation to identify homogeneous tokens in the image by identifying pixels that have the same two band, texton label, texton histogram label representation.

2. The method of claim 1 wherein the texton histogram label is generated by generating an histogram of texton label information.

3. The method of claim 1 comprising the further steps of:

identifying a texture region in the image; and

performing the converting step on the identified texture region of the image.

4. The method of claim 3 wherein the step of identifying a texture region in the image is carried out by generating a texture mask of the image.

5. An automated, computerized method for processing an image, comprising the steps of:

a computer executing the following steps:

converting a color band representation of each pixel in the image to a homogeneous representation by assigning spectral and spatial texture characteristics to each pixel in the image, the spectral and spatial characteristics including a representation of a texture primitive that is generated by a first clustering process and a representation of a spatial distribution of the texture primitive that is generated by a second clustering process; and

utilizing the homogeneous representation of spectral and spatial characteristics of a texture region in the image to identify homogeneous tokens in the image by identifying pixels that have the same spectral and spatial characteristics.